• AI Academy
  • Posts
  • 🤖 OpenAI Rushes o3-mini to Market Days After DeepSeek's Model

🤖 OpenAI Rushes o3-mini to Market Days After DeepSeek's Model

Plus: New Research Features and Europe's AI Regulations Begin

February 06, 2025

Hello AI Enthusiast,

This week in AI, OpenAI launched two major products in quick succession. First came o3-mini, a faster and cheaper reasoning model released just days after DeepSeek's announcement. They followed this with "deep research," a new tool that provides detailed, cited analysis by spending up to 30 minutes studying sources. Meanwhile, the EU's comprehensive AI regulations have officially begun. Let's dive in.

Before we dive into the news, join our next free webinar!

Transform your voice notes into polished content packages automatically! Join us on February 13th at 6 PM CET as Alessandro De Salve shows how his no-code AI prototype turns simple recordings into blog posts, images, and podcasts.

The Big Picture 🔊

OpenAI Releases o3-mini, a More Affordable Reasoning Model

OpenAI has launched o3-mini, its latest AI reasoning model designed for STEM problems. Unlike traditional language models, o3-mini fact-checks itself before providing results, making it more reliable for technical tasks. The model delivers answers 24% faster than its predecessor while making 39% fewer major mistakes. Available through ChatGPT and API access, o3-mini is priced 63% lower than previous versions, with different access tiers for free and paid users. Users can adjust the model's "reasoning effort" (low, medium, or high) to balance speed and accuracy according to their needs.

Andrea Mattiello
Andrea MattielloCM and Outdoor Fanatic

While o3-mini shows impressive improvements and lower costs, OpenAI might be creating a paradox of choice. ChatGPT now offers multiple models with subtle differences. Some can search the web but can't process images, others excel at STEM but struggle with general tasks

This rapid release, likely prompted by DeepSeek's competition, highlights how AI technology is advancing faster than user experience. The average user shouldn't need to choose between different AI models, the system should automatically select the best tool for each task.

OpenAI Launches 'Deep Research' for Complex Analysis

OpenAI has introduced 'deep research', a new AI agent designed for complex research tasks in fields like finance, science, policy, and engineering. Unlike regular ChatGPT interactions, this tool takes 5-30 minutes to analyze multiple sources thoroughly, providing fully documented responses with clear citations. Currently available to Pro users with a limit of 100 queries per month, it will roll out to other subscription tiers soon. The tool uses a specialized version of the o3 model optimized for web browsing and data analysis, though OpenAI acknowledges it may still make mistakes in distinguishing authoritative information from rumors.

Gianluca Belloni
Gianluca BelloniCMO and Marketing Nomad

While this feature is promising, it can still struggle to distinguish between authoritative information and rumors. Eliminating AI hallucinations is far more difficult than launching new features. The 30-minute wait time might make users more likely to trust the results without verification.

The timing is also telling, OpenAI launched this just two months after Google released a very similar feature in Gemini with the exact same name, highlighting the intense competition in web search.

EU's AI Act Begins Enforcement

The EU's comprehensive AI Act has begun rolling out its first phase of enforcement this week. Starting February 2, companies face strict regulations and potential fines. The Act bans several AI applications, including emotion recognition systems, facial recognition databases, and predictive policing. Additionally, organizations must ensure their employees achieve appropriate "AI literacy" levels based on their roles, with requirements ranging from basic awareness to advanced proficiency.

Helin Yontar
Helin YontarCPO and Polyglot

While the EU's intentions are good, there's a risk this could become just another box-ticking exercise. Companies might opt for the cheapest compliance solutions, like having employees watch basic videos without ensuring real understanding.

This highlights a key challenge: quick compliance versus meaningful education.

We're already seeing workers simply playing training videos in the background while multitasking. Having a training program that actually delivers results is crucial for long-term success. Our methodology is built on proven scientific principles that ensure lasting, practical skills. Want to learn more about "The Science Behind AI Academy's Trainings"?

Bits and Bobs 🗞️

  • Google has quietly launched its new flagship AI model, Gemini 2.0 Pro Experimental, for Gemini Advanced users.

  • The Gemini app has launched the Gemini 2.0 Flash model, offering faster responses and improved performance for tasks like brainstorming and writing.

  • Mistral AI has released a new model, Mistral Small 3, which matches the performance of larger AI models like Meta's Llama 3.3 70B while being more efficient and quicker, running almost 30% faster than GPT-4o Mini.

  • OpenAI suspects that DeepSeek may have improperly used data from its models to develop their own chatbot through a method called distillation.

  • DeepSeek faced a major security issue as researchers discovered an exposed database accessible with minimal effort, revealing user data and API keys.

  • The US Copyright Office recently stated that AI-generated content based solely on text prompts is not protected by copyright, as users lack sufficient control over the final output.

  • LinkedIn founder Reid Hoffman and cancer researcher Siddhartha Mukherjee have co-founded Manas AI, a startup focused on AI-powered drug discovery.

  • Perplexity now offers an uncensored version of DeepSeek R1, allowing users to access the AI model without China-based restrictions while storing data in the U.S.

  • Microsoft is introducing "Think Deeper," a feature in Copilot that allows it to tackle more complex questions by analyzing them from various perspectives.

  • Former Twitter co-founder Jack Dorsey has launched Goose, an open-source framework designed to streamline and automate tasks across multiple software platforms.

  • The Beatles' new track "Now and Then" was created using AI-assisted audio isolation techniques to enhance John Lennon’s muffled demo vocals.

On the Podcast 🎧

In the last episode of "I'm NOT a Robot," we talk with Faustine Cachera, an IP and data protection lawyer at Arendt & Medernach, about how AI is reshaping legal practice and what makes us irreplaceably human in an AI-powered world.

From Our Founder’s Channels 🤳

Watch Gianluca explain how o3-mini's pricing confirms his long-standing prediction that AI is becoming "the fastest depreciating asset in history".

@gianluca.mauro

OpenAI’s answer to deepseek is here #Ai #learnontiktok #artificialintelligence #business #machinelearning #product #ux #entrepreneur

LOLgorithms 😂

Karma’s got a sense of humor

That's a wrap on our newsletter! Before you go, here’s a quick recap of our offerings:

Catch you next week! 👋